Dario Lo Presti - Fotolia
When multiple virtual CPUs are allocated to the same VM, the hypervisor must wait for available processor cycles across multiple physical CPUs. If the workload is not designed for symmetric multi-processing (SMP), this behavior can actually reduce the workload’s apparent performance. The challenge here is not processor cycles -- there are probably plenty of processor cycles available across multiple CPUs. The challenge is scheduling (waiting for) time on multiple CPUs.
Monitoring can help administrators identify vCPU performance problems. For example, VMware vSphere’s CPU ready statistic (cpu.ready.summation) reports the time (milliseconds) that a VM was trying to execute without an available physical CPU. This value should ideally be small; as it grows, the VM is waiting longer for an available processor. By comparison, vSphere’s CPU usage statistic (cpu.usage.average) reports the percentage of time the CPU is used per host, per VM or per pool. A larger number indicates that the available processing resources are being more heavily utilized.
So if the CPU usage is fairly light, but the CPU ready (more accurately termed "wait") is unusually heavy, this may be a good indicator of multiple vCPU contention.
CPU limits and reservations are often not appropriate solutions to multiple vCPU provisioning problems. Limits put a cap on the maximum processing capacity of the CPU. Without limits, a VM can use 100% of the processor’s clock cycles, though CPUs are still shared between multiple VMs. Limits often cause unexpected (and unnecessary) VM performance problems for VMs that otherwise have proper vCPU provisioning. Reservations set a minimum amount of CPU processing capacity allocated to a VM -- even if the VM is not currently using that CPU capacity -- and can make that reserved capacity unavailable to other VMs which also results in potential resource shortages or performance issues.
For most VMs, allocating a single vCPU will provide the VM with access to a shared CPU core, and this usually provides enough processing capability to handle the VM’s needs. Although administrators can certainly allocate multiple virtual CPUs to the same VM, only VMs designed for SMP will really benefit, and other single-threaded workloads will often see a performance penalty.
Four mistakes that can hurt VM performanceEliminate IO bottlenecks to improve virtual machine performance
Does adding more virtual CPUs lead to better performance?
Dig Deeper on Server hardware and virtualization
Related Q&A from Stephen J. Bigelow
Just because software passes functional tests doesn't mean it works. Dig into stress, load, endurance and other performance tests, and their ... Continue Reading
Don't neglect form factor as part of your data center server selection. Instead, figure out what type of environment you need and learn which server ... Continue Reading
Learn how load balancing in the cloud differs from a traditional network traffic distribution, and explore the different services available from AWS,... Continue Reading